Abstract

In this work, we consider a class of structured regular LDPC codes, called Turbo-Structured LDPC (TS-LDPC). TS-LDPC codes outperform random LDPC codes and have much lower error floor at high Signal-to-Noise Ratio (SNR). In this thesis, Min-Sum (MS) algorithms are adopted in the decoding of TS-LDPC codes due to their low complexity in the implementation. We show that the error performance of the MS-based TS-LDPC decoder is comparable with the Sum-Product (SP) based decoder and the error floor property of TS-LDPC codes is preserved.

The TS-LDPC decoding algorithms can be performed by analog or digital circuitry. Analog decoders are preferred in many communication systems due to their potential for higher speed, lower power dissipation and smaller chip area compared to their digital counterparts. In this work, implementation of the (120, 75) MS-based TS-LDPC analog decoder is considered.

The decoder chip consists of an analog decoder heart, digital input and digital output blocks. These digital blocks are required to deliver the received signal to the
analog decoder heart and transfer the estimated codewords to the off-chip module. The analog decoder heart is an analog processor performing decoding on the Tanner graph of the code. Variable and check nodes are the main building blocks of analog decoder which are designed and evaluated. The check node is the most complicated unit in MS-based decoders. The minimizer circuit, the fundamental block of a check node, is designed to have a good trade-off between speed and accuracy. In addition, the structure of a high degree minimizer is proposed considering the accuracy, speed, power consumption and robustness against mismatch of the check node unit.

The measurement results demonstrate that the error performance of the chip is comparable with theory. The SNR loss at Bit-Error-Rate of 10−5 is only 0.2dB compared to the theory while information throughput is 750Mb/s and the energy efficiency of the decoder chip is 17pJ/b. It is shown that the proposed decoder outperforms the analog decoders that have been fabricated to date in the sense of error performance, throughput and energy efficiency. This decoder is the first analog decoder that has ever been implemented in a sub 100-nm technology and it improves
the throughput of analog decoders by a factor of 56. This decoder sets a new state-of-the-art in analog decoding.